Epidemiology and Risk Factors of Infection in Early Childhood

The Copenhagen Prospective Study on Asthma in Childhood 2000 (COPSAC₂₀₀₀) is a longitudinal birth cohort study of 411 children born to asthmatic mothers with a particular focus on asthma, allergy, and eczema as clinical outcomes along with a strong focus on infections as explanatory variables.²¹ Children born prematurely (<36 weeks of gestation), with severe congenital malformation or lower respiratory infection during the first month of life were excluded. In the first 3 years of life, the children attended the COPSAC₂₀₀₀ research clinic at 1 month of age and every 6 months thereafter for scheduled investigations as well as for acute care visits.²¹ The research clinic was de facto acting general practitioner for the children.

The study was approved by the Copenhagen Ethics Committee (KF01-289/96) and the Danish Data Protection Agency (2008-41-1754). Written and oral informed consent was obtained from both parents.

Infection burden was captured at the scheduled 6-monthly visit from birth until age 3 years and at additional acute care visits, where the COPSAC₂₀₀₀ pediatricians interviewed the parents about any illnesses, symptoms, duration, medication, and vaccinations since the last visit. Infections were classified according to the International Classification of Diseases, 10th Revision²² and stored in a designated database. When needed, the physician added additional clinical information. In case of missed visits, the parents were interviewed at the subsequent visit on infectious episodes. For the current study, the International Classification of Diseases, 10th Revision²² diagnoses were retrieved and grouped as follows:

• Upper respiratory tract infections (URTIs): common cold, tonsillitis, pharyngitis, otitis media, and croup;

• Lower respiratory tract infections (LRTIs): pneumonia and bronchiolitis;

• Gastrointestinal infections (GIs): GI, diarrhea, and vomiting; and

• Isolated fever and other infections.

We analyzed the overall incidence of infections (the 4 groups combined) and the incidence rates of groups I, II, and III separately. Group VI comprised a heterogeneous group of infections without respiratory or gastrointestinal symptoms and was not analyzed as a separate outcome. Details on diagnoses can be found in Supplemental Information (Supplemental Table 5).

Source data validation was performed to evaluate the quality of data as previously published.²³ Records from the COPSAC₂₀₀₀ database were compared with information recorded by the children’s general practitioner and revealed ˃90% completeness and no important influence from socioeconomics or concurrent asthma.

A total of 84 environmental and constitutional risk factors were collected during the first 3 years of life. These risk factors include information on demographics, third trimester exposures, maternal infections during pregnancy, newborn characteristics, neonatal biomarkers, postnatal exposures, day care attendance, diet, indoor environment, and genetics^21,24,–29 (Supplemental Table 6).

First, we applied a quasi-Poisson regression model estimating unadjusted incidence rate ratios (IRRs) for the following 18 variables suspected to be associated with infection burden: sex,¹⁸ birth weight and length, mode of delivery (vaginal or caesarean),³⁰ paternal asthma, maternal smoking during pregnancy,^13,14 older siblings,^3,5,12 cats or dogs in the home,³¹ household income,^17,18 the mother’s occupation and level of education,¹⁷ duration of breastfeeding,^9,10 day care attendance^1,6,–8,11 (age at introduction, number of children, and m² per child in the day care), and ETS assessed by nicotine concentration in the child’s hair at ages 1 and 3 years.³² These variables were selected a priori on the basis of literature to reduce the risk of multiple testing. The quasi-Poisson model was used to account for overdispersion in the data. Thereafter, we performed an adjusted quasi-Poisson regression analysis including risk factors with a P value ˂.20, estimating adjusted incidence rate ratios (aIRRs). In the adjusted analysis, some variables were closely correlated (ie, smoking during pregnancy and ETS, birth weight and birth length, and socioeconomic variables), and, in these cases, only 1 variable was included for adjustment to avoid collinearity (see Supplemental Information for clarification). We performed a stratified analysis on the basis of a diagnosis of asthma at any time point before age 3 years.

Secondly, we applied an unsupervised data-driven sparse principal component analysis (SPCA)^33,34 to explore common underlying patterns from the entire set of variables from the COPSAC₂₀₀₀ database without a priori hypotheses on association to infections (ie, a total of 84 descriptive covariates) (Supplemental Table 6). SPCA is a modification of the multivariate technique principal component analysis (PCA). Further details can be found in the Supplemental Information. The SPCA patterns were used as input variables in a forward stepwise Poisson regression for the prediction of incidence of infections.

The quasi-Poisson regression analyses were performed in SAS (SAS Institute, Inc, Cary, NC) version 9.3 and R version 2.12.0. The SPCA analyses were performed in MatLab version R2016b by the algorithm available on http://models.life.ku.dk/sparsity.³⁴ 

Children who missed 2 or more successive 6-monthly scheduled visits were excluded from the analysis leaving 334 (81%) of the 411 children eligible for analysis. Children in the study group were significantly more often born by caesarean delivery, exposed to a cat at home, attended more crowded day cares, and came from families with higher household incomes but were less exposed to maternal smoking during pregnancy (Table 1). A total of 327 (98%) children were fully vaccinated according to the national immunization program (Supplemental Table 7).

A total of 5009 infections were reported among the 334 children (Fig 1A), yielding a median incidence rate of 14 infections per child (mean 15; range 2–43; interquartile range (IQR) 10–18). Respiratory tract infections were most frequent with a median of 10 episodes per child, corresponding to 71% of all infections (9 episodes per child for URTI and 1 episode per child for LRTI) (Table 2).

With Figure 1B, we show the incidence rates of infections at ages 0 to 3 years, illustrating an incidence peak around age 1 year with a slight decline in the third year of life. Figure 1C reveals the seasonal variation in incidence rates with respiratory tract infections being more common in the winter, whereas fever and gastroenteritis had less seasonal variation.

Table 2 reveals the mean durations of infectious episodes for each category. Median duration of all infectious episodes was 6 days (IQR 3–8), which decreased with increasing age (IRR per year 0.94 [0.90–0.97]; P = .001). The same inverse association between disease duration and age was seen for URTI (IRR 0.96 [0.92–1.00]; P = .057), LRTI (IRR 1.10 [0.87–0.94]; P = .02), and GI (IRR 0.84 [0.73–0.96]; P = .01), whereas the duration of isolated fever increased with age (IRR 1.10 [1.02–1.19]; P = .02).

The median number of days with infection throughout the 3-year study period was 94 days (IQR 64–132), in which the majority of days were caused by URTI (62 days [IQR 40–97]) (Table 2).

Antibiotics were administered in 24.9% of the infectious episodes (87.7% of LRTI episodes, 23.1% of URTI episodes, 12.5% of isolated fever episodes, and 10.2% of GI episodes). The most frequently used drug was amoxicillin (59.4%) followed by penicillin (27.9%) (Fig 2, Supplemental Table 8).

The only risk factors significantly related to the overall incidence of infections was crowding in day care, measured as m² per child in the day care (aIRR per IQR: 0.96 [0.92–0.99]; P = .04), meaning that children in day care centers with m² per child in the lower quartile had 4% fewer infections than those in the upper quartile (Table 3). Also, the total number of children in the day care was associated with increased incidence of infections (aIRR 1.09 [1.02–1.16]; P = .01), meaning that children attending day care with the number of children in the upper quartile experienced 9% more infections than those in the lower quartile.

The same associations were seen for crowding in day care, in which the m² per child was inversely associated with incidence of URTI (aIRR per IQR 0.95 [0.91–1.00]; P = .048), and the total number of children in the day care was associated with increased incidence of URTI (aIRR 1.11 [1.03–1.20]; P = .01). We observed no other significant associations.

Risk factors associated with incidence of LRTI were caesarean delivery (aIRR 1.49 [1.12–1.99]; P = .01), maternal smoking during pregnancy (aIRR 1.66 [1.18–2.23]; P < .005), and the presence of older siblings (aIRR 1.54 [1.19–2.01]; P < .005), whereas an older age at the introduction to day care was inversely associated with LRTI incidence (aIRR per IQR 0.77 [0.65–0.91]; P < .005). A subanalysis on the effect of the age of the youngest older siblings revealed a tendency toward the sibling effect declining with the age of the youngest older sibling; however, this decline was not statistically significant (P = .40). Male sex was associated with borderline significance (aIRR 1.29 [1.00–1.68]; P = .052).

No significant associations were found.

Children diagnosed with persistent wheeze and/or asthma before age 3 years had the same infection burden as the remaining children, except they had a higher number of LRTIs (IRR 2.97 [2.36–3.75]; P < .001). Excluding children with persistent wheeze and/or asthma yielded the same risk factor profile for LRTIs (Supplemental Table 9).

The 84 environmental and constitutional covariates were decomposed into 11 underlying latent risk factor patterns called components. Supplemental Figure 3 reveals the key covariates driving each component and the percentage of the total variance that each component accounts for. In total, the 11 components could only be used to explain 44% of the overall data variation.

Table 4 reveals the association between the components and incidence of infections. Only component 1 was significantly associated with the overall incidence of infections (IRR per IQR 0.91 [0.86–0.97]; P = .003), but the component could only be used to explain 8.4% of the variation in the data. Described in component 1 was a pattern primarily driven by measures of ETS, combined with socioeconomics, indoor air pollution, and pets (Supplemental Fig 3). Unexpectedly, lower levels of ETS were associated with increased incidence of infections. Component 1 was also associated with incidence of URTI (IRR per IQR 0.93 [0.86–0.99]; P = .03), and there was a trend toward an inverse association between component 1 and incidence of LRTI (IRR per IQR 1.15 [0.99–1.34]; P = .06), meaning that higher levels of ETS increased risk of LRTI.

Component 4 was significantly associated with LRTI (IRR per IQR 0.83 [0.70–0.98]; P = .03). This component was primarily driven by breastfeeding, both the duration of solely breastfeeding and any breastfeeding. Component 8 was significantly associated with both LRTI (IRR per IQR 0.78 [0.66–0.92]; P = .004) and GI (IRR per IQR 0.85 [0.74–0.98]; P = .03). This component was primarily driven by furred pets and 17q21 polymorphisms. Excluding asthmatic children from the analysis did not alter the results.

This prospective Danish birth cohort study revealed that otherwise healthy children experienced a median of 14 simple infectious episodes throughout the first 3 years of life (mean 15; IQR 10–18; range 2–43) with substantial variation in frequency between individuals.

Only crowding in day care had a significant but modest influence on the overall incidence of infections and incidence of URTI, whereas LRTI incidence was associated with maternal smoking, caesarean delivery, older siblings, and early day care attendance. When including 84 environmental and constitutional covariates with an explorative SPCA approach, we were only able to describe 8.4% of the large variance in infection frequency. With these findings, we suggest that host factors are the major determinants of infection susceptibility in early childhood.

The strength of our study is the close longitudinal prospective clinical surveillance at the COPSAC₂₀₀₀ clinical research unit that included 3 years of follow-up and a total observation of 365 730 days. The study is a single-center study with 6 monthly assessments conducted by experienced study pediatricians who examined the children and obtained their clinical history that was supported by daily diary cards. Through this, data capture was secured and the risk of misclassification was reduced. We previously validated our data with records from children’s general practitioners²³ and showed good data capture with sensitivity ˃90%.

Highly detailed assessment of the exposome is a key feature of the COPSAC₂₀₀₀ study, including information on a wide range of environmental and constitutional risk factors collected prospectively, combining interview data with objective measurements.²¹ 

It is an advantage that, with our study, we combine traditional inferential statistical methods with an unsupervised data-driven analysis. The quasi-Poisson regression analysis reveals if a single risk factor is significantly associated with higher infection rate but does not reveal whether a pattern comprising several risk factors carries the relevant information. Many exposure variables are correlated, and a multiparametric approach is therefore needed, such as the SPCA, which can be used to handle an underlying correlation structure. Furthermore, the SPCA approach overcomes the issue of multiple testing in unidirectional analyses. Using this method, we were able to consider the high number of exposure variables available in our cohort study without a priori selection of risk factors.

It is a limitation that the selection of collected variables is driven by the cohort’s focus on atopic diseases, and a cohort study in which childhood infections are primarily addressed would possibly include other variables. However, few such studies have been conducted,^2,3,5,11 and all include less risk factors compared with this study.

The generalizability of our findings can be questioned because of the high risk of asthma in the cohort. Children at high risk of developing asthma and wheezy symptoms may experience more infections than nonwheezing children,³⁵ which could lead to an overestimation of population disease incidence in this cohort. Still, we would not expect this to affect the observed wide variation of infections within the cohort or the influence of risk factors. Furthermore, a sensitivity analysis excluding children with persistent wheeze and/or asthma yielded similar results.

With this unparalleled comprehensive risk factor analysis, in which we use both traditional statistics and SPCA pattern recognition analysis, we confirm the lack of evident and reproducible associations between simple childhood infections and exposure-related risk factors. These findings reveal that host factors rather than the exposome account for the variation in frequency of infections in early childhood.

The authors of previous studies have reported that day care attendance is associated with increased risk of respiratory tract infection especially during the first years of life,^2,7,36 but the authors of studies with longitudinal follow-up have questioned whether the increased symptom burden continues later in childhood.^8,37,38 We confirmed a modest influence from attendance to crowded day care facilities, reflected as a 9% increase in overall incidence of infections and an 11% increase in URTIs among children attending day care with crowding in the upper quartile compared with the lower. Likewise, we saw a 4% reduction in overall incidence of infections and a 5% reduction in URTIs per IQR of available space per child in the day care. The effect of day care is presumably caused by an increased disease transmission.

Children with older siblings have been suggested to suffer from more respiratory infections than first-born children,³⁹ but it has also been proposed that children with older siblings experience an earlier immune maturation¹² and a subsequent improved resistance against infections later in childhood. We found that having siblings at home increased the risk of LRTIs by 54% but had no influence on the incidence of overall infections, URTIs, or GIs.

Boys are expected to have an increased risk of respiratory infections, particularly LRTI.^39,40 We found a tendency toward increased incidence of LRTIs, but there was no apparent association between sex and other types of infections.

ETS, estimated both by maternal smoking during pregnancy and by nicotine level in the child’s hair, was associated with increased frequency of LRTIs, and, although not statistically significant, the SPCA analysis supported the positive association between ETS and incidence of LRTI, which aligns with other reports.¹³ We previously showed that children exposed to tobacco smoke in utero exhibit a deficit in lung function at birth,^41,42 which could make them more prone to LRTIs. There was no association between URTI and ETS in the univariate analysis, and, surprisingly, the SPCA pointed toward a protective effect of ETS, which could be a false discovery. The lack of association is in fact in line with other prospective studies,² supporting the role of ETS as a trigger of lower respiratory symptoms, such as cough and wheeze, but not enhancing the susceptibility to infections per se.

Cesarean delivery was associated with a 49% increased incidence of LRTI. The authors of a previous epidemiologic birth cohort study conducted in Norway on the basis of questionnaires filled out by 37 101 mothers were unable to find a significant association,³⁰ but, although not statistically significant, all of their results pointed toward an increased risk of recurrent LRTI within the first 3 years of life after caesarean delivery. The mechanism behind this association is unclear. It could be a reflection of other risk factors influencing the likelihood of a caesarean delivery, but it might be that children born by vaginal delivery are exposed to a diverse microbiological flora from the birth canal⁴³ and that subsequent colonization of gut and airways alters immune modulation⁴⁴ and susceptibility to LRTI.⁴⁵ 

Pets at home, namely dogs, have been found protective of respiratory tract infection in early childhood.³¹ We did not find univariate association between pet exposure and incidence of infections, but, with the SPCA, we pointed toward an increased overall incidence of infections for children with furred pets in the home, both cat and dog. Furthermore, pet exposure in combination with certain wheeze-related genotypes (17q21) was related to incidence of both LRTI and GI, but, in this setting, dog exposure increased the number of LRTIs and GIs, whereas cat exposure seemed to protect. The lack of clear and reproducible results underlines the complexity of these environmental exposures.

The burden of infections could be associated with vaccination status. However, vaccination rates were high in our study, and we were unable to address influence from vaccines. The children were vaccinated according to the Danish immunization schedule at the time of the study (diphtheria-tetanus toxoids-pertussis-polio, Haemophilus influenzae type b, and measles-mumps-rubella; Supplemental Table 7). Since then, vaccination for several other pathogens, including Pneumococcus, Meningococcus, rotavirus, and varicella-zoster virus are being implemented in many countries. It is a subject for authors of future studies to explore whether these vaccines reduce the incidence of childhood infections in industrialized countries, either through specific protection or through nonspecific immune-modulatory effects.⁴⁶ 

The most notable finding is that so few of so many suspected risk factors contributed substantially to the variation in incidence of infections, particularly the overall incidence and incidence of URTI and GI. This lack of evident and reproducible associations between simple childhood infections and various exposure-related risk factors is in agreement with other studies,^5,19 suggesting that unidentified host factors are the major determinants of the highly variable infection burden in young children. Therefore, authors of future studies should search for alternative explanations of disease patterns presumably through a systems biology omics approach including other areas such as functional immunology, genetics, dietary patterns, microbiome, and inflammatory responses early in life.

Children experienced a median of 14 simple infections during the first 3 years of life, with 71% being respiratory infections. Individual variation in disease frequency for all infections and URTIs was associated with crowding in day care, and LRTI was associated with day care attendance, ETS, caesarean delivery, duration of breastfeeding, and having older siblings at home. However, these risk factors explained only a small fraction (8.4%) of the interindividual variation in incidence of infections, suggesting that host factors are the major determinants.

aIRR

adjusted incidence rate ratio

COPSAC₂₀₀₀

Copenhagen Prospective Studies on Asthma in Childhood 2000

ETS

environmental tobacco exposure

GI

gastrointestinal infection

IQR

interquartile range

IRR

incidence rate ratio

LRTI

lower respiratory tract infection

PCA

principal component analysis

SPCA

sparse principal component analysis

URTI

upper respiratory tract infection

We thank the children and families of the COPSAC₂₀₀₀ cohort study for all their support and commitment. We acknowledge and appreciate the unique efforts of the COPSAC₂₀₀₀ research team.